PartialTypeSignatures trigger bogus "unbound implicit parameter" error

GHC seems to forget about in-scope implicit parameters when there's a hole in the type signature. Given

{-# LANGUAGE ImplicitParams, PartialTypeSignatures #-}

module Bad where

f1 :: (?loc :: Int, _) => Int
f1 = ?loc

f2 :: (?loc :: Int) => _
f2 = ?loc

GHC incorrectly reports the following errors.

% ghc -fforce-recomp Bad.hs
[1 of 1] Compiling Bad              ( Bad.hs, Bad.o )

Bad.hs:8:6:
    Unbound implicit parameter (?loc::Int)
      arising from a use of implicit parameter ‘?loc’
    In the expression: ?loc
    In an equation for ‘f1’: f1 = ?loc

Bad.hs:11:1:
    Occurs check: cannot construct the infinite type:
      _ ~ (?loc::Int) => _
    When checking that ‘f2’ has the specified type
      f2 :: (?loc::Int) => _
    Probable cause: the inferred type is ambiguous

Bad.hs:11:6:
    Unbound implicit parameter (?loc::_)
      arising from a use of implicit parameter ‘?loc’
    Relevant bindings include f2 :: _ (bound at Bad.hs:11:1)
    In the expression: ?loc
    In an equation for ‘f2’: f2 = ?loc

?loc is very clearly bound by both f1 and f2's signature. f2 additionally reports what looks to me to be a bogus occurs-check error; we know from the theta that ?loc :: Int so the hole should be solved for Int.

I suspect this is all related to the fact that PartialTypeSignatures triggers tcPolyInfer instead of tcPolyCheck.

I think this is the same underlying bug as #10846 (closed).

Trac metadata

Trac field	Value
Version	7.10.2
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tbug Trac import labels

I see that commit 1e041b73 addresses this issue and fixes the 'unbound implicit parameter' error (thanks!!), but ghc still refuses to compile f2.

% ./inplace/bin/ghc-stage2 Bad.hs
[1 of 1] Compiling Bad              ( Bad.hs, Bad.o )

Bad.hs:9:1: error:
    • Could not deduce: ?loc::t
      from the context: ?loc::Int
        bound by the inferred type for ‘f2’:
                   (?loc::Int) => t
        at Bad.hs:9:1-9
    • When checking that the inferred type
        f2 :: forall t. (?loc::t) => t
      is as general as its (partial) signature
        f2 :: forall t. (?loc::Int) => t

I think this error is still bogus, because ghc should not have inferred

        f2 :: forall t. (?loc::t) => t

in the first place. It should have inferred

        f2 :: (?loc::Int) => Int

which is a valid instantiation of the partial signature, and so should be OK.

mentioned in commit 28035c09

closed

Excellent point, thank you!

Fixed.

Simon

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed
Test case	- → partial-sigs/should_compile/T11016

changed milestone to %8.0.1

added Pnormal label